st: Models that don't impose IIA?

[Adrian Gonzalez-Gonzalez <gonzalez-gonzale.1@osu.edu>]

Dear Statalisters

 

I have the following problem:

-My dependent variable is categorical

-It is not ordered

-The explanatory variables are characteristics of the individuals making the choices, and not of the choices per se.

 

The easiest model to estimate is a multinomial logit, but it has the IIA restriction.

 

I cannot estimate a Nested Logit (in order to get over the IIA) because I don’t have independent variables that characterized the choices.

Some other more flexible models are: Mixed Logit (aka: Random Parameters Logit or Error-Component Logit) or Multinomial Probit, but they cannot be estimated using Stata (I will be extremely happy to hear refutations on this)

 

Is there any other model that does not impose the IIA restriction and that can be estimated with Stata?  Any thoughts of the topic?

 

Adrian

 

 

PS: Thank you for the postings on the tests of IIA under mlogit, and the existence of mprobit for Stata.

 

 

Adrian Gonzalez

Ph.D Candidate

Agricultural, Environmental and Development Economics

The Ohio State University